ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-24 23:24:13 +00:00
Code Issues Packages Projects Releases Wiki Activity

q8_KV: slightly faster gemv on Zen4

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-02-18 10:40:50 +02:00
parent 7f4ec2f964
commit 1ecea16f63
1 changed files with 27 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -6245,47 +6245,41 @@ template <int nrc_y>
static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataInfo& info, int nrc_x) {
GGML_ASSERT(nrc_x%8 == 0);
GGML_ASSERT(n%32 == 0);
#ifndef HAVE_FANCY_SIMD
__m256i qx[2];
__m256i acc[nrc_y] = {};
float dy[nrc_y];
#ifdef HAVE_FANCY_SIMD
int32_t sy[nrc_y];
#else
__m256i sx[2];
auto m1 = _mm256_set1_epi16(1);
#endif
__m256i qx[4];
__m256i sx[4];
__m256i acc[nrc_y] = {};
float dy[nrc_y];
const int8_t * q8y[nrc_y];
for (int iy = 0; iy < nrc_y; ++iy) {
auto dptr = (const float *)info.src1_row(iy);
dy[iy] = dptr[0];
#ifdef HAVE_FANCY_SIMD
auto iptr = (const int32_t *)(dptr+1);
sy[iy] = -127*iptr[0];
#endif
q8y[iy] = (const int8_t *)(dptr + 2);
}
for (int ix = 0; ix < nrc_x; ++ix) {
auto dx = (const float *)((const char *)vx + ix*bx);
auto q8x = (const int8_t *)(dx + 2);
for (int i = 0; i < n/128; ++i) {
for (int j = 0; j < 4; ++j) {
qx[j] = _mm256_loadu_si256((const __m256i *)q8x + 4*i + j);
sx[j] = _mm256_sign_epi8(qx[j], qx[j]);
}
for (int iy = 0; iy < nrc_y; ++iy) {
for (int j = 0; j < 4; ++j) {
#ifdef HAVE_FANCY_SIMD
acc[iy] = _mm256_dpbusd_epi32(acc[iy], sx[j], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + 4*i + j), qx[j]));
#else
auto dot = _mm256_maddubs_epi16(sx[j], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + 4*i + j), qx[j]));
acc[iy] = _mm256_add_epi32(acc[iy], _mm256_madd_epi16(m1, dot));
#endif
}
}
}
for (int i = 2*(n/128); i < n/64; ++i) {
for (int i = 0; i < n/64; ++i) {
for (int j = 0; j < 2; ++j) {
#ifdef HAVE_FANCY_SIMD
qx[j] = _mm256_add_epi8(_mm256_loadu_si256((const __m256i *)q8x + 2*i + j), _mm256_set1_epi8(127));
#else
qx[j] = _mm256_loadu_si256((const __m256i *)q8x + 2*i + j);
sx[j] = _mm256_sign_epi8(qx[j], qx[j]);
#endif
}
for (int iy = 0; iy < nrc_y; ++iy) {
for (int j = 0; j < 2; ++j) {
#ifdef HAVE_FANCY_SIMD
acc[iy] = _mm256_dpbusd_epi32(acc[iy], sx[j], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + 2*i + j), qx[j]));
acc[iy] = _mm256_dpbusd_epi32(acc[iy], qx[j], _mm256_loadu_si256((const __m256i *)q8y[iy] + 2*i + j));
#else
auto dot = _mm256_maddubs_epi16(sx[j], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + 2*i + j), qx[j]));
acc[iy] = _mm256_add_epi32(acc[iy], _mm256_madd_epi16(m1, dot));
@@ -6294,11 +6288,15 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI
}
}
if (int i = 2*(n/64); i < n/32) {
#ifdef HAVE_FANCY_SIMD
qx[0] = _mm256_add_epi8(_mm256_loadu_si256((const __m256i *)q8x + i), _mm256_set1_epi8(127));
#else
qx[0] = _mm256_loadu_si256((const __m256i *)q8x + i);
sx[0] = _mm256_sign_epi8(qx[0], qx[0]);
#endif
for (int iy = 0; iy < nrc_y; ++iy) {
#ifdef HAVE_FANCY_SIMD
acc[iy] = _mm256_dpbusd_epi32(acc[iy], sx[0], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + i), qx[0]));
acc[iy] = _mm256_dpbusd_epi32(acc[iy], qx[0], _mm256_loadu_si256((const __m256i *)q8y[iy] + i));
#else
auto dot = _mm256_maddubs_epi16(sx[0], _mm256_sign_epi8(_mm256_loadu_si256((const __m256i *)q8y[iy] + i), qx[0]));
acc[iy] = _mm256_add_epi32(acc[iy], _mm256_madd_epi16(m1, dot));
@@ -6307,7 +6305,11 @@ static void mul_mat_q8_KV_q8_KV_1(int n, const void * vx, size_t bx, const DataI
}
for (int iy = 0; iy < nrc_y; ++iy) {
auto sumi = hsum_i32_8(acc[iy]);
info.store(ix, iy, dx[0]*dy[2*iy+0]*sumi);
#ifdef HAVE_FANCY_SIMD
info.store(ix, iy, dx[0]*dy[iy]*(sumi+sy[iy]));
#else
info.store(ix, iy, dx[0]*dy[iy]*sumi);
#endif
acc[iy] = _mm256_setzero_si256();
}
}